Race To The Future
Thomas More - Digital Innovation · 2024
Race to the Future:
AI-Driven RC Car in Action
The challenge
An annual tradition within Digital Innovation: building an RC car and training it to race autonomously. Every year the Digital Innovation students take on the Industrial Engineers from KU Leuven in an exciting AI race. This year the challenge was avoiding two strategically placed cones, one of which was positioned to be extra difficult.
How we approached it
We started by converting an RC car and installing a Jetson Nano as its AI brain. With a wide-angle camera (160° FOV) and DonkeyCar, an open-source platform for self-driving cars, we were able to train a model to navigate on its own.
Training data was collected by driving the car manually with a PlayStation 4 controller. The camera recorded images of the track while steering and throttle values were saved. We collected 110,000 images, of which 20,000 were removed because they were unclear or captured crashes. In the end we trained our model on a dataset of 90,000 images.
The result
After weeks of testing and optimising, our AI was able to race the car autonomously and avoid obstacles. During the competition against the KU Leuven students, our model proved its strength: we managed to beat them and take the win!
What we learned
This project gave us not only technical skills in AI and machine learning, but also insight into teamwork and problem solving. The competition added an extra challenge and the motivation to make our AI as good as it could be.